Skip to main content

Table 4 Comparison of the three ensemble methods for both data sets with respect to the micro-F measure

From: Large-scale online semantic indexing of biomedical articles via an ensemble of multi-label classification models

Micro-F measure

       

Data set

MetaLabeler

SVM Tuned

SVM Vanilla

LLDA

Improve micro-F

Improve F [13]

MULE

A

       
 

  

0.58546

0.58127

0.58705

 

  

0.58601

0.58260

0.58734

   

0.55522

0.52144

0.55675

  

0.57246

0.54166

0.57458

 

0.58695

0.55836

0.58919

B

       
 

  

0.50136

0.49445

0.50435

 

  

0.50144

0.49329

0.50522

   

0.44159

0.42726

0.44304

  

0.46247

0.45685

0.45868

 

0.50058

0.49227

0.50353

  1. “Improve micro-F” is the initial version of MULE, without the statistical test. “Improve-F” is the method proposed by [13]. A symbol suggests that the difference with the best performing model is statistically significant with a z-test and a significance level of 0.05